| In today’s information age,application software is everywhere,and software quality issues come with it.Software defects are the antithesis of software quality,making the security and reliability of software face serious threats.Therefore,how to accurately and efficiently predict defects in software modules has become an urgent problem to be solved.Software defect prediction is one of the important ways of software quality assurance,and has become a very important research topic in software engineering in recent years.However,software defect prediction with the help of machine learning is still facing the following problems in practice: imbalanced data class,low prediction accuracy and so on.How to effectively solve the above problems has become the current research hotspot in related fields.This thesis systematically studies the problems of imbalanced data processing in software defect prediction from the perspective of data sampling and ensemble learning.First,we improve the traditional SMOTE algorithm and get a weighted complexity based on SMOTE algorithm WCP-SMOTE,and apply it to software defect prediction;Second,we propose a new undersampling algorithm WCP-Under Sampler based on weighted complexity.We compare the WCP-Under Sampler algorithm with the classical sampling method,and study the effects of WCP-Under Sampler algorithm on the performance of software defect prediction model;Finally,we combined the data sampling method with ensemble learning method,in order to increase the diversity of base classifiers,an ensemble learning algorithm E＿RHFS based on self-help sampling and hybrid feature selection is proposed and applied to software defect prediction.The main work of this thesis can be summarized as follows:（1）An SMOTE Algorithm Based on Weighted Complexity and Its Application in Software Defect PredictionTo deal with the problem of class imbalance in software defect prediction,it is studies the shortcomings of the traditional SMOTE oversampling method,improve the traditional SMOTE algorithm,and propose the WCP-SMOTE algorithm.WCP-SMOTE algorithm uses the sample complexity to assist in the synthesis of new samples,so as to provide an effective solution for the class imbalance problem.Comparing WCP-SMOTE algorithm with traditional oversampling methods,experiments on multiple Promise datasets show that using WCP-SMOTE to deal with imbalanced data can obtain better defect prediction performance.（2）An Under-sampling Algorithm Based on Weighted Complexity and Its Application in Software Defect PredictionTo deal with the problem of class imbalance in software defect prediction,propose a new undersampling algorithm based on weighted complexity（ie WCP-Under Sampler algorithm）,which avoids deleting some samples valuable to model prediction in the process of under sampling,so as to provide an effective solution for the class imbalance problem.Comparing WCP-Under Sampler algorithm with traditional sampling methods,experiments on multiple Promise datasets show that WCP-Under Sampler algorithm can obtain better defect prediction performance on KNN and CART classifiers.（3）An Ensemble Learning Algorithm Based on Self-help Sampling and Hybrid Feature Selection,with an Application to Software Defect PredictionEnsemble learning is widely used in software defect prediction,but the existing methods still have the following problems: the diversity of base classifiers is difficult to guarantee,and the prediction accuracy is low.To solve the above problems,this thesis proposes an ensemble learning algorithm E＿RHFS based on self-help sampling and hybrid feature selection,and uses it to predict software defects.To deal with the problem of class imbalance,the data are balanced using sampling techniques before executing the E＿RHFS algorithm.On the balanced data set,E＿RHFS algorithm first uses self-help sampling technology to generate multiple sampling sets,and uses granularity decision entropy and random subspace to select hybrid features for each sampling set;Second,a base classifier is constructed on each sample set selected by hybrid features;Finally,all base classifiers are integrated by voting,so as to obtain the final integrated classifier for software defect prediction.Experiments on multiple Promise datasets show that better defect prediction performance can be obtained by combining E＿RHFS algorithm with data sampling algorithm. |
